From fc6eead7c1e2e5376c25d2795d4539fdacbc0648 Mon Sep 17 00:00:00 2001 From: John Stultz Date: Mon, 22 May 2017 17:20:20 -0700 Subject: [PATCH 1/4] time: Clean up CLOCK_MONOTONIC_RAW time handling Now that we fixed the sub-ns handling for CLOCK_MONOTONIC_RAW, remove the duplicitive tk->raw_time.tv_nsec, which can be stored in tk->tkr_raw.xtime_nsec (similarly to how its handled for monotonic time). Cc: Thomas Gleixner Cc: Ingo Molnar Cc: Miroslav Lichvar Cc: Richard Cochran Cc: Prarit Bhargava Cc: Stephen Boyd Cc: Kevin Brodsky Cc: Will Deacon Cc: Daniel Mentz Tested-by: Daniel Mentz Signed-off-by: John Stultz --- arch/arm64/kernel/vdso.c | 6 ++-- include/linux/timekeeper_internal.h | 4 +-- kernel/time/timekeeping.c | 45 ++++++++++++++++------------- 3 files changed, 29 insertions(+), 26 deletions(-) diff --git a/arch/arm64/kernel/vdso.c b/arch/arm64/kernel/vdso.c index d0cb007fa482..7492d9009610 100644 --- a/arch/arm64/kernel/vdso.c +++ b/arch/arm64/kernel/vdso.c @@ -220,10 +220,8 @@ void update_vsyscall(struct timekeeper *tk) if (!use_syscall) { /* tkr_mono.cycle_last == tkr_raw.cycle_last */ vdso_data->cs_cycle_last = tk->tkr_mono.cycle_last; - vdso_data->raw_time_sec = tk->raw_time.tv_sec; - vdso_data->raw_time_nsec = (tk->raw_time.tv_nsec << - tk->tkr_raw.shift) + - tk->tkr_raw.xtime_nsec; + vdso_data->raw_time_sec = tk->raw_sec; + vdso_data->raw_time_nsec = tk->tkr_raw.xtime_nsec; vdso_data->xtime_clock_sec = tk->xtime_sec; vdso_data->xtime_clock_nsec = tk->tkr_mono.xtime_nsec; vdso_data->cs_mono_mult = tk->tkr_mono.mult; diff --git a/include/linux/timekeeper_internal.h b/include/linux/timekeeper_internal.h index f7043ccca81c..0a0a53daf2a2 100644 --- a/include/linux/timekeeper_internal.h +++ b/include/linux/timekeeper_internal.h @@ -51,7 +51,7 @@ struct tk_read_base { * @clock_was_set_seq: The sequence number of clock was set events * @cs_was_changed_seq: The sequence number of clocksource change events * @next_leap_ktime: CLOCK_MONOTONIC time value of a pending leap-second - * @raw_time: Monotonic raw base time in timespec64 format + * @raw_sec: CLOCK_MONOTONIC_RAW time in seconds * @cycle_interval: Number of clock cycles in one NTP interval * @xtime_interval: Number of clock shifted nano seconds in one NTP * interval. @@ -93,7 +93,7 @@ struct timekeeper { unsigned int clock_was_set_seq; u8 cs_was_changed_seq; ktime_t next_leap_ktime; - struct timespec64 raw_time; + u64 raw_sec; /* The following members are for timekeeping internal use */ u64 cycle_interval; diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index b602c48cb841..0454bfa24353 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c @@ -72,6 +72,10 @@ static inline void tk_normalize_xtime(struct timekeeper *tk) tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; tk->xtime_sec++; } + while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) { + tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift; + tk->raw_sec++; + } } static inline struct timespec64 tk_xtime(struct timekeeper *tk) @@ -285,12 +289,14 @@ static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) /* if changing clocks, convert xtime_nsec shift units */ if (old_clock) { int shift_change = clock->shift - old_clock->shift; - if (shift_change < 0) + if (shift_change < 0) { tk->tkr_mono.xtime_nsec >>= -shift_change; - else + tk->tkr_raw.xtime_nsec >>= -shift_change; + } else { tk->tkr_mono.xtime_nsec <<= shift_change; + tk->tkr_raw.xtime_nsec <<= shift_change; + } } - tk->tkr_raw.xtime_nsec = 0; tk->tkr_mono.shift = clock->shift; tk->tkr_raw.shift = clock->shift; @@ -619,9 +625,6 @@ static inline void tk_update_ktime_data(struct timekeeper *tk) nsec = (u32) tk->wall_to_monotonic.tv_nsec; tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); - /* Update the monotonic raw base */ - tk->tkr_raw.base = timespec64_to_ktime(tk->raw_time); - /* * The sum of the nanoseconds portions of xtime and * wall_to_monotonic can be greater/equal one second. Take @@ -631,6 +634,11 @@ static inline void tk_update_ktime_data(struct timekeeper *tk) if (nsec >= NSEC_PER_SEC) seconds++; tk->ktime_sec = seconds; + + /* Update the monotonic raw base */ + seconds = tk->raw_sec; + nsec = (u32)(tk->tkr_raw.xtime_nsec >> tk->tkr_raw.shift); + tk->tkr_raw.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); } /* must hold timekeeper_lock */ @@ -672,7 +680,6 @@ static void timekeeping_update(struct timekeeper *tk, unsigned int action) static void timekeeping_forward_now(struct timekeeper *tk) { u64 cycle_now, delta; - u64 nsec; cycle_now = tk_clock_read(&tk->tkr_mono); delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); @@ -684,10 +691,13 @@ static void timekeeping_forward_now(struct timekeeper *tk) /* If arch requires, add in get_arch_timeoffset() */ tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift; - tk_normalize_xtime(tk); - nsec = clocksource_cyc2ns(delta, tk->tkr_raw.mult, tk->tkr_raw.shift); - timespec64_add_ns(&tk->raw_time, nsec); + tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult; + + /* If arch requires, add in get_arch_timeoffset() */ + tk->tkr_raw.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_raw.shift; + + tk_normalize_xtime(tk); } /** @@ -1373,19 +1383,18 @@ int timekeeping_notify(struct clocksource *clock) void getrawmonotonic64(struct timespec64 *ts) { struct timekeeper *tk = &tk_core.timekeeper; - struct timespec64 ts64; unsigned long seq; u64 nsecs; do { seq = read_seqcount_begin(&tk_core.seq); + ts->tv_sec = tk->raw_sec; nsecs = timekeeping_get_ns(&tk->tkr_raw); - ts64 = tk->raw_time; } while (read_seqcount_retry(&tk_core.seq, seq)); - timespec64_add_ns(&ts64, nsecs); - *ts = ts64; + ts->tv_nsec = 0; + timespec64_add_ns(ts, nsecs); } EXPORT_SYMBOL(getrawmonotonic64); @@ -1509,8 +1518,7 @@ void __init timekeeping_init(void) tk_setup_internals(tk, clock); tk_set_xtime(tk, &now); - tk->raw_time.tv_sec = 0; - tk->raw_time.tv_nsec = 0; + tk->raw_sec = 0; if (boot.tv_sec == 0 && boot.tv_nsec == 0) boot = tk_xtime(tk); @@ -2011,15 +2019,12 @@ static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset, *clock_set |= accumulate_nsecs_to_secs(tk); /* Accumulate raw time */ - tk->tkr_raw.xtime_nsec += (u64)tk->raw_time.tv_nsec << tk->tkr_raw.shift; tk->tkr_raw.xtime_nsec += tk->raw_interval << shift; snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift; while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) { tk->tkr_raw.xtime_nsec -= snsec_per_sec; - tk->raw_time.tv_sec++; + tk->raw_sec++; } - tk->raw_time.tv_nsec = tk->tkr_raw.xtime_nsec >> tk->tkr_raw.shift; - tk->tkr_raw.xtime_nsec -= (u64)tk->raw_time.tv_nsec << tk->tkr_raw.shift; /* Accumulate error between NTP and clock interval */ tk->ntp_error += tk->ntp_tick << shift; From 369adf04d80a7e179b9ea6d74cc01c233f142f47 Mon Sep 17 00:00:00 2001 From: John Stultz Date: Fri, 12 May 2017 10:58:18 -0700 Subject: [PATCH 2/4] time: Add warning about imminent deprecation of CONFIG_GENERIC_TIME_VSYSCALL_OLD CONFIG_GENERIC_TIME_VSYSCALL_OLD was introduced five years ago to allow a transition from the old vsyscall implementations to the new method (which simplified internal accounting and made timekeeping more precise). However, PPC and IA64 have yet to make the transition, despite in some cases me sending test patches to try to help it along. http://patches.linaro.org/patch/30501/ http://patches.linaro.org/patch/35412/ If its helpful, my last pass at the patches can be found here: https://git.linaro.org/people/john.stultz/linux.git dev/oldvsyscall-cleanup So I think its time to set a deadline and make it clear this is going away. So this patch adds warnings about this functionality being dropped. Likely to be in v4.15. Cc: Thomas Gleixner Cc: Ingo Molnar Cc: Miroslav Lichvar Cc: Richard Cochran Cc: Prarit Bhargava Cc: Marcelo Tosatti Cc: Paul Mackerras Cc: Anton Blanchard Cc: Benjamin Herrenschmidt Cc: Tony Luck Cc: Michael Ellerman Cc: Fenghua Yu Signed-off-by: John Stultz --- kernel/time/timekeeping.c | 1 + 1 file changed, 1 insertion(+) diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index 0454bfa24353..cedafa008de5 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c @@ -516,6 +516,7 @@ static void halt_fast_timekeeper(struct timekeeper *tk) } #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD +#warning Please contact your maintainers, as GENERIC_TIME_VSYSCALL_OLD compatibity will disappear soon. static inline void update_vsyscall(struct timekeeper *tk) { From 7a5de5512296fd2ab7a497e4a576196b1f046e78 Mon Sep 17 00:00:00 2001 From: Miroslav Lichvar Date: Tue, 13 Jun 2017 12:57:07 +0200 Subject: [PATCH 3/4] kselftests: timers: Fix inconsistency-check to not ignore first timestamp When the first timestamp in the list of clock readings was later than the second timestamp and all other timestamps were in order, the inconsistency was not reported because the index of the out-of-order timestamp was equal to the default value. Cc: Thomas Gleixner Cc: Ingo Molnar Cc: Miroslav Lichvar Cc: Richard Cochran Cc: Prarit Bhargava Cc: Stephen Boyd Cc: Shuah Khan Signed-off-by: Miroslav Lichvar Signed-off-by: John Stultz --- tools/testing/selftests/timers/inconsistency-check.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tools/testing/selftests/timers/inconsistency-check.c b/tools/testing/selftests/timers/inconsistency-check.c index caf1bc9257c4..74c60e8759a0 100644 --- a/tools/testing/selftests/timers/inconsistency-check.c +++ b/tools/testing/selftests/timers/inconsistency-check.c @@ -118,7 +118,7 @@ int consistency_test(int clock_type, unsigned long seconds) start_str = ctime(&t); while (seconds == -1 || now - then < seconds) { - inconsistent = 0; + inconsistent = -1; /* Fill list */ for (i = 0; i < CALLS_PER_LOOP; i++) @@ -130,7 +130,7 @@ int consistency_test(int clock_type, unsigned long seconds) inconsistent = i; /* display inconsistency */ - if (inconsistent) { + if (inconsistent >= 0) { unsigned long long delta; printf("\%s\n", start_str); From 767392565a3e618950fe1a5ff1ba11295f6332f4 Mon Sep 17 00:00:00 2001 From: Miroslav Lichvar Date: Fri, 9 Jun 2017 13:43:17 +0200 Subject: [PATCH 4/4] kselftests: timers: Add test for frequency step This test checks the response of the system clock to frequency steps made with adjtimex(). The frequency error and stability of the CLOCK_MONOTONIC clock relative to the CLOCK_MONOTONIC_RAW clock is measured in two intervals following the step. The test fails if values from the second interval exceed specified limits. Cc: Thomas Gleixner Cc: Ingo Molnar Cc: Miroslav Lichvar Cc: Richard Cochran Cc: Prarit Bhargava Cc: Stephen Boyd Cc: Shuah Khan Signed-off-by: Miroslav Lichvar Signed-off-by: John Stultz --- tools/testing/selftests/timers/Makefile | 5 +- tools/testing/selftests/timers/freq-step.c | 268 +++++++++++++++++++++ 2 files changed, 271 insertions(+), 2 deletions(-) create mode 100644 tools/testing/selftests/timers/freq-step.c diff --git a/tools/testing/selftests/timers/Makefile b/tools/testing/selftests/timers/Makefile index 5fa1d7e9a915..5801bbefbe89 100644 --- a/tools/testing/selftests/timers/Makefile +++ b/tools/testing/selftests/timers/Makefile @@ -1,6 +1,6 @@ BUILD_FLAGS = -DKTEST CFLAGS += -O3 -Wl,-no-as-needed -Wall $(BUILD_FLAGS) -LDFLAGS += -lrt -lpthread +LDFLAGS += -lrt -lpthread -lm # these are all "safe" tests that don't modify # system time or require escalated privileges @@ -8,7 +8,7 @@ TEST_GEN_PROGS = posix_timers nanosleep nsleep-lat set-timer-lat mqueue-lat \ inconsistency-check raw_skew threadtest rtctest TEST_GEN_PROGS_EXTENDED = alarmtimer-suspend valid-adjtimex adjtick change_skew \ - skew_consistency clocksource-switch leap-a-day \ + skew_consistency clocksource-switch freq-step leap-a-day \ leapcrash set-tai set-2038 set-tz @@ -24,6 +24,7 @@ run_destructive_tests: run_tests ./change_skew ./skew_consistency ./clocksource-switch + ./freq-step ./leap-a-day -s -i 10 ./leapcrash ./set-tz diff --git a/tools/testing/selftests/timers/freq-step.c b/tools/testing/selftests/timers/freq-step.c new file mode 100644 index 000000000000..e8c61830825a --- /dev/null +++ b/tools/testing/selftests/timers/freq-step.c @@ -0,0 +1,268 @@ +/* + * This test checks the response of the system clock to frequency + * steps made with adjtimex(). The frequency error and stability of + * the CLOCK_MONOTONIC clock relative to the CLOCK_MONOTONIC_RAW clock + * is measured in two intervals following the step. The test fails if + * values from the second interval exceed specified limits. + * + * Copyright (C) Miroslav Lichvar 2017 + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of version 2 of the GNU General Public License as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License for more details. + */ + +#include +#include +#include +#include +#include + +#include "../kselftest.h" + +#define SAMPLES 100 +#define SAMPLE_READINGS 10 +#define MEAN_SAMPLE_INTERVAL 0.1 +#define STEP_INTERVAL 1.0 +#define MAX_PRECISION 100e-9 +#define MAX_FREQ_ERROR 10e-6 +#define MAX_STDDEV 1000e-9 + +struct sample { + double offset; + double time; +}; + +static time_t mono_raw_base; +static time_t mono_base; +static long user_hz; +static double precision; +static double mono_freq_offset; + +static double diff_timespec(struct timespec *ts1, struct timespec *ts2) +{ + return ts1->tv_sec - ts2->tv_sec + (ts1->tv_nsec - ts2->tv_nsec) / 1e9; +} + +static double get_sample(struct sample *sample) +{ + double delay, mindelay = 0.0; + struct timespec ts1, ts2, ts3; + int i; + + for (i = 0; i < SAMPLE_READINGS; i++) { + clock_gettime(CLOCK_MONOTONIC_RAW, &ts1); + clock_gettime(CLOCK_MONOTONIC, &ts2); + clock_gettime(CLOCK_MONOTONIC_RAW, &ts3); + + ts1.tv_sec -= mono_raw_base; + ts2.tv_sec -= mono_base; + ts3.tv_sec -= mono_raw_base; + + delay = diff_timespec(&ts3, &ts1); + if (delay <= 1e-9) { + i--; + continue; + } + + if (!i || delay < mindelay) { + sample->offset = diff_timespec(&ts2, &ts1); + sample->offset -= delay / 2.0; + sample->time = ts1.tv_sec + ts1.tv_nsec / 1e9; + mindelay = delay; + } + } + + return mindelay; +} + +static void reset_ntp_error(void) +{ + struct timex txc; + + txc.modes = ADJ_SETOFFSET; + txc.time.tv_sec = 0; + txc.time.tv_usec = 0; + + if (adjtimex(&txc) < 0) { + perror("[FAIL] adjtimex"); + ksft_exit_fail(); + } +} + +static void set_frequency(double freq) +{ + struct timex txc; + int tick_offset; + + tick_offset = 1e6 * freq / user_hz; + + txc.modes = ADJ_TICK | ADJ_FREQUENCY; + txc.tick = 1000000 / user_hz + tick_offset; + txc.freq = (1e6 * freq - user_hz * tick_offset) * (1 << 16); + + if (adjtimex(&txc) < 0) { + perror("[FAIL] adjtimex"); + ksft_exit_fail(); + } +} + +static void regress(struct sample *samples, int n, double *intercept, + double *slope, double *r_stddev, double *r_max) +{ + double x, y, r, x_sum, y_sum, xy_sum, x2_sum, r2_sum; + int i; + + x_sum = 0.0, y_sum = 0.0, xy_sum = 0.0, x2_sum = 0.0; + + for (i = 0; i < n; i++) { + x = samples[i].time; + y = samples[i].offset; + + x_sum += x; + y_sum += y; + xy_sum += x * y; + x2_sum += x * x; + } + + *slope = (xy_sum - x_sum * y_sum / n) / (x2_sum - x_sum * x_sum / n); + *intercept = (y_sum - *slope * x_sum) / n; + + *r_max = 0.0, r2_sum = 0.0; + + for (i = 0; i < n; i++) { + x = samples[i].time; + y = samples[i].offset; + r = fabs(x * *slope + *intercept - y); + if (*r_max < r) + *r_max = r; + r2_sum += r * r; + } + + *r_stddev = sqrt(r2_sum / n); +} + +static int run_test(int calibration, double freq_base, double freq_step) +{ + struct sample samples[SAMPLES]; + double intercept, slope, stddev1, max1, stddev2, max2; + double freq_error1, freq_error2; + int i; + + set_frequency(freq_base); + + for (i = 0; i < 10; i++) + usleep(1e6 * MEAN_SAMPLE_INTERVAL / 10); + + reset_ntp_error(); + + set_frequency(freq_base + freq_step); + + for (i = 0; i < 10; i++) + usleep(rand() % 2000000 * STEP_INTERVAL / 10); + + set_frequency(freq_base); + + for (i = 0; i < SAMPLES; i++) { + usleep(rand() % 2000000 * MEAN_SAMPLE_INTERVAL); + get_sample(&samples[i]); + } + + if (calibration) { + regress(samples, SAMPLES, &intercept, &slope, &stddev1, &max1); + mono_freq_offset = slope; + printf("CLOCK_MONOTONIC_RAW frequency offset: %11.3f ppm\n", + 1e6 * mono_freq_offset); + return 0; + } + + regress(samples, SAMPLES / 2, &intercept, &slope, &stddev1, &max1); + freq_error1 = slope * (1.0 - mono_freq_offset) - mono_freq_offset - + freq_base; + + regress(samples + SAMPLES / 2, SAMPLES / 2, &intercept, &slope, + &stddev2, &max2); + freq_error2 = slope * (1.0 - mono_freq_offset) - mono_freq_offset - + freq_base; + + printf("%6.0f %+10.3f %6.0f %7.0f %+10.3f %6.0f %7.0f\t", + 1e6 * freq_step, + 1e6 * freq_error1, 1e9 * stddev1, 1e9 * max1, + 1e6 * freq_error2, 1e9 * stddev2, 1e9 * max2); + + if (fabs(freq_error2) > MAX_FREQ_ERROR || stddev2 > MAX_STDDEV) { + printf("[FAIL]\n"); + return 1; + } + + printf("[OK]\n"); + return 0; +} + +static void init_test(void) +{ + struct timespec ts; + struct sample sample; + + if (clock_gettime(CLOCK_MONOTONIC_RAW, &ts)) { + perror("[FAIL] clock_gettime(CLOCK_MONOTONIC_RAW)"); + ksft_exit_fail(); + } + + mono_raw_base = ts.tv_sec; + + if (clock_gettime(CLOCK_MONOTONIC, &ts)) { + perror("[FAIL] clock_gettime(CLOCK_MONOTONIC)"); + ksft_exit_fail(); + } + + mono_base = ts.tv_sec; + + user_hz = sysconf(_SC_CLK_TCK); + + precision = get_sample(&sample) / 2.0; + printf("CLOCK_MONOTONIC_RAW+CLOCK_MONOTONIC precision: %.0f ns\t\t", + 1e9 * precision); + + if (precision > MAX_PRECISION) { + printf("[SKIP]\n"); + ksft_exit_skip(); + } + + printf("[OK]\n"); + srand(ts.tv_sec ^ ts.tv_nsec); + + run_test(1, 0.0, 0.0); +} + +int main(int argc, char **argv) +{ + double freq_base, freq_step; + int i, j, fails = 0; + + init_test(); + + printf("Checking response to frequency step:\n"); + printf(" Step 1st interval 2nd interval\n"); + printf(" Freq Dev Max Freq Dev Max\n"); + + for (i = 2; i >= 0; i--) { + for (j = 0; j < 5; j++) { + freq_base = (rand() % (1 << 24) - (1 << 23)) / 65536e6; + freq_step = 10e-6 * (1 << (6 * i)); + fails += run_test(0, freq_base, freq_step); + } + } + + set_frequency(0.0); + + if (fails) + ksft_exit_fail(); + + ksft_exit_pass(); +}